a fresh look at hpc - hpc advisory council · page 4 huawei technologies co., ltd. maximize...

20
www.huawei.com HUAWEI TECHNOLOGIES CO., LTD. A Fresh Look at HPC HUAWEI TECHNOLOGIES Francis Lam Director, Product Management

Upload: others

Post on 15-Mar-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

www.huawei.comHUAWEI TECHNOLOGIES CO., LTD.

A Fresh Look at HPCHUAWEI TECHNOLOGIES

Francis LamDirector, Product Management

Page 2: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 2 HUAWEI TECHNOLOGIES CO., LTD.

WORLD CLASS HPC SOLUTIONS TODAY170+

Countries 2016 Revenue

16

R&D Centers

36Joint Innovation

Centers

79,000

R&D Engineers

Standalone Compute

Node 1P-32PModular HPC

Systems

NVMe SSD

HPC storage

Big Datastorage

Network FabricModular & Container

Data center

$74.8B

of Revenue in R&D

14.2%

Huawei FusionServer

OceanStor CloudEngine

Reduce Complexity

More Performance / $

Design for Growth

HPC Private Cloud

》》

Petascale System

Direct Liquid Cooling

Workload Optimization

Ecosystem Partnership

》》

Page 3: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 3 HUAWEI TECHNOLOGIES CO., LTD.

300+ World Records For Computing Performance

Flexible Density Optimized

Rack Mount Servers

High Performance Blades

Large Scale Deployment

SERVERS FOR HIGH PERFORMANCE COMPUTING

RH5885

RH8100

RH1288/2288

KunLun

16/32S System

RH5288

NVMe SSD card

Acceleration

GPGPU/Phi

E9000

X6800

Page 4: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 4 HUAWEI TECHNOLOGIES CO., LTD.

MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD

MAXIMIZE PERFORMANCE FOR

INDIVIDUAL WORKLOAD

Flexible, modular architecture

Multiple innovative form factors

Deep optimization with hardware

acceleration

Super Fat nodes

ADAPT TO CHANGE

FUTURE PROOF

Solve problem using emerging

hardware technologies

Single HPC cluster and storage

system for both traditional HPC

MPI workload and Hadoop

Leverage flexibility and

expandability of cloud technology

HUAWEI ADVANTAGES

MORE COMPUTE LESS SPACE

LOWER POWER CONSUMPTION

End-to-end energy efficient design

HVDC

High Ambient Temperature ~40oC

Direct Liquid Cooling ~ 84% coverage

Tight integration with Huawei data

center infrastructure

SDS

Big Data

SDI

Page 5: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 5 HUAWEI TECHNOLOGIES CO., LTD.

Energy Efficient Server Design

Bottom line of Greener IT

Higher Temperature

Up to 8%↓ WW DC power consumption w/ 5ºC↑

HVDC

Up to 9-15% conversion efficiency

improvement

Free Air Cooling

Up to 20-70% cooling saving

In-Memory Computing

Up to 90+% less power

consumption than HDD-based

SSD Storage

Up to 60+% less power consumption than

HDD-based

Air Containment

Up to 30%↑ in cooling efficiency

Green IT Reduces Energy Bill & CO2 Emission, Extends DC Life, Lowers TCO

Hot Water

Eliminate Chiller Free cooling

Right-sized Power

~20% better power & cooling utilization

END-TO-END GREEN HPC DESIGN

Hardware Acceleration

~ 2X+ Performance Per Watt

Page 6: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 6 HUAWEI TECHNOLOGIES CO., LTD.

CPU Load XH620 V3(X6800) 1U Server

Watt Saving Per

Node

% Power Saving Per

Node

100% 303 313 10 3%

90% 267 283 16 6%

80% 225 235 10 4%

70% 191 206 15 7%

60% 166 183 17 9%

50% 152 164 12 7%

40% 139 151 12 8%

30% 127 138 11 8%

20% 114 125 11 9%

10% 101 112 11 10%

0% 61 73 11.5 16%

X6800

Vs.

Workload:SPECpower2008,Power Meter:WT210,Thermometer:Digi Watchport/H

1U Server

POWER EFFICIENT FORM FACTORS

Page 7: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 7 HUAWEI TECHNOLOGIES CO., LTD.

HUAWEI FUSIONSERVER LIQUID COOLING SYSTEM

• FusionServer LCS is composed of Liquid Cooling Rack and external CDU

• CPU, Memory and VRD are cooled directly by up to 45 ℃ water

• Chiller is optional, cooling PUE < 1.1

• Industry leading serviceability

Page 8: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 8 HUAWEI TECHNOLOGIES CO., LTD.

Scalability

I/O

Storage

Memory

bandwidth

Memory

capacity

CPU

Workload Characteristics

WORKLOAD OPTIMIZED HIGH PERFORMANCE BLADES

Maximize CPU compute density

Maximize Memory Per Core

Maximize Memory BandwidthMaximize Local Storage Per Node

Maximize I/O, Hardware Acceleration

Page 9: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 9 HUAWEI TECHNOLOGIES CO., LTD.

E9000 HIGH PERFORMANCE BLADE SYSTEMC

ha

ss

is E9000 chassis

Chassis

Co

mp

ute

no

de

CH121 V3

Sw

itch

mo

du

le

Switch module

CH220 V3CH140 V3

Half-width 2-socket

compute node

High density

Large memory

capacity

Adopts a modular design for computing, storage, switching, power supply, and cooling.

12-U-high chassis, providing 8 full-width or 16 half-width slots.

Supports next 3 generations of high-performance Intel CPUs.

Supports next-decade network technology evolution.

Compute node

CH222 V3 CH226 V3 CH242 V3

Half-width 2*2-socket twin

compute nodes

Super high density

Outstanding computing

capability

Full-width I/O

expansion compute

node

Large memory

6*PCIe slots

Full-width storage node

Large memory

15*2.5'' HDDs/SSDs

Full-width storage node

Large memory

6*3.5'' HDDs

Full-width 4-socket compute node

Outstanding computing capability

(E7 v2, E7 v3)

Superb storage and I/O expansion

capability

GE 10GE/FCoE converged Ethernet IB FDR/EDR Multi-plane switch module8G/16G FC 40GE

CX110 CX310 CX311CX116 pass-

through

CX317/CX318

pass-through

CX611 10GE+FC

CX912

CX111 CX210

8G FC

GE+FC

CX915

CX710CX220

16G FC

CH225 V3

Full-width storage node

All-flash

12*NVMe SSDs

Page 10: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 10 HUAWEI TECHNOLOGIES CO., LTD.

X6800 HIGH DENSITY RACK MOUNT SERVERS

XH628 V3

XH622 V3

XH620 V3

High Density Compute Node

Big Data Storage Node

Hardware acceleration Node

Page 11: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 11 HUAWEI TECHNOLOGIES CO., LTD.

BROAD RANGE OF NVME SSD SERVERS

4 ╳ NVMe RH1288 V3

2 ╳ NVMe XH620 V3

12 ╳ NVMe RH2288H V3

24 ╳ NVMe RH2288H V34 ╳ NVMe 5288 V3

4 ╳ NVMe RH2288(H) V3

NVMe SSD Disk

2 ╳ NVMe CH121 V3

12 ╳ NVMe CH225 V3

4 ╳ NVMe CH242 V3

4 ╳ NVMe RH5885(H) V3

8 ╳ NVMe RH8100 V3

Blade Server

X6800(4U8)

6 ╳ NVMe XH321 V3

High-Density Server

X6000(2U4)

E9000

Page 12: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 12 HUAWEI TECHNOLOGIES CO., LTD.

FUSIONSERVER X6000 — HIGH-DENSITY SERVER

Front View Rear View

24 NVMe SSDs or SAS/SATA HDDsUnified

management

network port

platinum AC PSUs

(hot-swappable, 1+1

redundancy)

2 PCIe slots

LOM service port

Node management port

Universal connector port

Page 13: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 13 HUAWEI TECHNOLOGIES CO., LTD.

ACCELERATE HIGH MEMORY BANDWIDTH WORKLOAD

WITH FAT NODESRFD’s tNavigator has superior scalability but memory bandwidth constrained

Huawei and RFD collaborated to design and benchmark a new cluster system with Intel E7 4-Socket high memory bandwidth node

Excellent benchmark results

Page 14: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 14 HUAWEI TECHNOLOGIES CO., LTD.

Large Memory Fat Node

Huawei FusionServer RH8100 V3

12TB RAM

8 Intel Xeon E7 v3 processors

High RAS capability

High serviceability

SnapVX: http://snap.stanford.edu/snapvx

Page 15: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 15 HUAWEI TECHNOLOGIES CO., LTD.

Unified Storage V3

5300, 5500, 5600, 5800, 6800Block + File

Mass Data Storage

UDS

Object for Near line & Offline

All-in-one Mass Data Storage

OceanStor 9000

File + Object + Analysis + Archiving

P15P22P1P8

P15P22P1P8

P15P22P1P8

P15P22P1P8

P15P22P1P8

P15P22P1P8

Unified block & file storage For global cloud data centers:

Decentralized & distributed

architecture & highly efficient

global data sharing

All-in-One storage solution

simplifying data lifecycle

management

HUAWEI OCEANSTOR STORAGE PORTFOLIO

Virtualization, SSD, Distributed Storage, Software Defined Storage

All flash Storage

Dorado 2100G2 & 5100Block

Full flash storage: best ratio

of performance/price

Page 16: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 16 HUAWEI TECHNOLOGIES CO., LTD.

SDI Enabled Server

SDI Controller (SoC)

Software Defined Deep Learning

LEAP FORWARD WITH HUAWEI INNOVATIONS

Enabling HPC Cloud

Open Telekom Cloud

FusionInsight

Big Data

FusionSphere

Cloud OS

ManageOne

Management software

FusionStorage

Software Defined Storage Pool

〉〉〉〉

〉〉〉〉

〉〉〉〉

〉〉〉〉

Performance Optimized Spark

Big Data Acceleration

Page 17: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Open Telekom Cloud 06.02.2017 17

NEWSLETTER ON CERN HPC PROJECTAccelerating science: Open Telekom Cloud in use at CERN

http://www.telekom.com/media/enterprise-solutions/307124

Page 18: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 18 HUAWEI TECHNOLOGIES CO., LTD.

HPC CLOUD CASE STUDY

Key Requirements

HPC resource pool

• 1000VMs(≧ 4vCPU /8GB MEM

/100GB local disk)

• No CPU/RAM overcommitment

• KV Performance: ≦ 1.2s/event

Low latency block disk

• 500TB block disk

• From VM to block storage

roundtrip access RTT≦5ms

High bandwidth network

• Bandwidth from in-house OpenStack

data center to VPC on OTC: 10Gbps

• NAT to internet for VMs = 1:1

• 20TB/per day data transmission from

in-house OpenStack Cloud to OTC

1000VMs on OTC at the first stage. Potentially 10000 VMs on OTC in the future

EBSVPC

EIP

Direct ConnectECS

Page 19: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 19 HUAWEI TECHNOLOGIES CO., LTD.

Manufacturing CAE/CFD Education/Research/Supercomputing

HUAWEI HPC MOMENTUM

Chip Design & Manufacturing

Oil & Gas ExplorationEnergy Production & Distribution Digital Media

Page 20: A Fresh Look at HPC - HPC Advisory Council · Page 4 HUAWEI TECHNOLOGIES CO., LTD. MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD Flexible, modular

Page 20 HUAWEI TECHNOLOGIES CO., LTD.

HUAWEI HPC SOLUTIONS

MAXIMIZE ACCELERATE ADAPT